gabrgomes / simple-user-api

simple-user-api showcase

This is a project created to showcase a simple python api running on AWS using ECS. The app implement api calls to upsert and get users on a MongoDB database. This is by no means intended for production usage.

The project is publicly available on https://github.com/gabrgomes/simple-user-api

Very simplified diagram

flowchart TB;
web((Internet)) -->|80| lb{Applicaton LB}
subgraph "cluster-app (ECS)"
App1 & App2 & App3
end
subgraph "DNS (Route 53)"
sd(mongo.app_zone)
end
subgraph "cluster-db (ECS)"
subgraph service
db[(MongoDB)]
end
end
lb -->|8000| App1 & App2 & App3
App1 & App2 & App3 -.->|mongo.app_zone:27017| sd
sd -.- service-discovery -.- service

Loading

Running locally

Requirements:

• docker >= 20.10.18

# Building the app
docker compose build 

# run containers on background
docker compose up -d

# Api doc will be available on http://localhost:8000/docs

# stop the containers
docker compose down

Running on AWS

Requirements:

• aws credentials configured (ref)
• terraform >= 1.3.2

# enter terraform directory
cd terraform

# install the requirements
terraform init

# see the modifications 
terraform plan

# provision application resources
terraform apply

# Api doc url will be shown on the output var app_url.

# remove application resources
terraform destroy

Variables

var	default	description
region_name	us-east-1	AWS region name
app_name	simple-user-api	Name of the application to be used to create resources.
app_image	public.ecr.aws/f9q5q0t9/simple-user-api:latest	Application image url
app_container_port	8000	Application port
db_container_port	27017	MongoDB port
max_capacity	3	Maximum number of application tasks
min_capacity	1	Minimum number of application tasks
cpu_threshold	80	Cpu % threshold for autoscaling
mem_threshold	80	Memory % threshold for autoscaling

Considerations

For simplicity this project has some questionable choices that should be mentioned:

• The terraform module provision all resources on the default VPC.
• The database is implemented as simple MongoDB instance with no authentication and no persistence.
• The task definition is pointing to a image on a public repository that will eventualy stop working at any time. In this case you will need to build and publish the image to your own registry.
• The containers in ECS are configured with public IP adresses to allow image pull from public repositories. Ideally you could use private repos on ECR with the necessary security group configurations or use NAT for outgoing connections to the internet.

Monitoring

The terraform code creates a simple dashboard on Cloud Watch with cpu and memory usage for the services.

TODO

• App
• Dockerfile
• Docker compose
• Deploy AWS
• HA
• Monitoring
• Autoscaling

Further improvements

• Create VPC and subnets
• Use Internet gateway
• Create pipeline to build image, push to ecr and update task definition
• Enable tracing with AWS Distro for OpenTelemetry